A problem for some learning classical guitar players (myself included) is excessive movement of the right hand. Ideally the hand itself is extremely stable, with all the movement in the fingers, and a jumpy hand makes it harder to play accurately. Various ways to address this are used, for example a mirror, exercises, and keeping contact with unplayed strings.
I thought it would be interesting to try doing some monitoring and real-time feedback using an accelerometer. I've been using a spatial phidget 1049 to experiment with the idea. It's easily light enough to sit on the back of my hand, having been sewn onto an elastic strap. I've tried two programs: one makes a bleep when the acceleration along one axis exceeds a threshold; the other records the acceleration for a period so that I can analyse it and see whether I can achieve a reduction over time.
So far this is promising. The threshold for immediate feedback has to be set much lower for my teacher than for me, so it seems like it's measuring something relevant. The acceleration graphs show the beat of the music nicely, so I could also look at my rhythm that way if I wanted to - at least until I've got my hand totally stable.
I wrote most of the code in Java. Working with the Phidgets API in the event-driven paradigm was straightforward. (On the other hand, learning to use the javax.sound API was, as so often with Java, far more fiddly than it should have been.) I called the Java classes from Matlab to simplify plotting graphs, passing parameters, and writing files. (Matlab and Java are very integrated.)
Maybe there are other ways to use these sensors for feedback in learning physical skills?